247 / 2024-03-14 12:27:39

Graph Regularization A general framework for multi-label Hashing

Hashing,Multi-label,graph,regularization

Conventional hashing methods aim to generate the compact binary codes under the assumption that the data for hashing is usually associated with a single label. However, in the datasets of NUS-WIDE, an image is often with a set of labels simultaneously. To deal with this issue, some researchers proposed to convert the multi-label problem into a serial of single label problem, and then learn the optimal hashing function for  multimedia retrieval. Although promising performance can be achieved, the label correlation information is ignored. In this paper, we propose a general framework based on graph regularization for multi-label hashing. On one hand, we design a general framework for binary codes learning. On the other hand, a graph Laplacian term includes the graph structure based on the sample space and the hypergraph structure based on the label space is embedded into the framework for addressing Multi-label hashing. Experiments on two standard image databases show that the performance of the proposed method is superior to other methods.